A neural network to pulmonary embolism aided diagnosis with a feature selection approach

Objectives: The purpose of this study was to build a backpropagation neural network (BNN) as a computer-aided diagnostic model based on selected input features for predicting pulmonary embolism (PE). Methods: We retrospectively reviewed 102 PE suspicious patient records with demographic characteristics, clinical symptoms, blood gas, D-dimer, and wells score. A logistic regression (LR) model was employed to extracted important predictive features, which used as inputs to the BNN model. The BNN was trained and tested using leave-one-out method and then the area under the receiver operating characteristic (ROC) curves was calculated to measure the performance. Results: The variables extracted from logistic regression enabled the BNN model achieved an Az =0.889±0.042 compare to the non-selected BNN model with Az=0.838±0.052. Conclusion: The results indicate that the logistic regression method and the backpropagation neural network, particularly when used in combination, can produce better predictive models than BNN alone. The features such as D-dimer, PO2, and history of deep vein thrombosis (DVT) or PE are beneficial for the differential diagnosis of PE. The Computer-aided diagnosis (CAD) system can help physicians to detect or exclude PE in the clinical practice, and it is a new promising method of diagnosing pulmonary embolism.